Stata Tip 40: Taking Care of Business

Abstract

Daily data are often generated by nondaily processes: for instance, financial markets are closed on weekends and holidays. Stata’s time-series date schemes ([U] 24.3 Timeseries dates) allow for daily data, but gaps in time series may be problematic. A model that uses lags or differences will lose several observations every time a gap appears, discarding many of the original data points. Analysis of “business-daily” data often proceeds by assuming that Monday follows Friday, and so on. At the same time, we usually want data to be placed on Stata’s time-series calendar so that useful tools such as the tsline graph will work and label data points with readable dates; see [TS] tsline. At a recent Stata Users Group presentation in Boston, David Drukker spoke to this point. His solution: generate two date variables, one containing the actual calendar dates, another numbering successive available observations consecutively. The former variable (caldate )i stsset (see [TS] tsset) when the calendar dates are to be used, whereas the latter (seqdate )i stsset when statistical analyses are to be performed. We download daily data on the 3-month U.S. Treasury bill rate with Drukker’s freduse command (Drukker 2006) and retain the August 2005–present data for analysis. (We can also view the data graphically with tsline.)